Abstract
Deep Echo State Networks (DeepESNs) recently extended the applicability of Reservoir Computing (RC) methods towards the field of deep learning. In this paper we study the impact of constrained reservoir topologies in the architectural design of deep reservoirs, through numerical experiments on several RC benchmarks. The major outcome of our investigation is to show the remarkable effect, in terms of predictive performance gain, achieved by the synergy between a deep reservoir construction and a structured organization of the recurrent units in each layer. Our results also indicate that a particularly advantageous architectural setting is obtained in correspondence of DeepESNs where reservoir units are structured according to a permutation recurrent matrix.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
- 2.
- 3.
- 4.
With the only exception of the case \(L=3\), where the first two layers contained 167 units and the last one contained 166 units.
- 5.
Performance differences between DeepESN with permutation topology and all the other architectures are confirmed by Wilcoxon rank-sum test performed at 1% significance level on all the tasks (with the only exceptions of the comparisons with ESN using chain topology on Laser, and ESN using permutation topology on MG30).
References
Bacciu, D., Bongiorno, A.: Concentric ESN: assessing the effect of modularity in cycle reservoirs. In: 2018 International Joint Conference on Neural Networks (IJCNN), pp. 1–8. IEEE (2018)
Boedecker, J., Obst, O., Mayer, N.M., Asada, M.: Studies on reservoir initialization and dynamics shaping in echo state networks. In: Proceedings of the 17th European Symposium on Artificial Neural Networks (ESANN), pp. 227–232. d-side publi. (2009)
Farkaš, I., Bosák, R., Gergel’, P.: Computational analysis of memory capacity in echo state networks. Neural Netw. 83, 109–120 (2016). https://doi.org/10.1016/j.neunet.2016.07.012
Farmer, J.D.: Chaotic attractors of an infinite-dimensional dynamical system. Physica D 4(3), 366–393 (1982). https://doi.org/10.1016/0167-2789(82)90042-2
Gallicchio, C.: Short-term memory of deep RNN. In: Proceedings of the 26th European Symposium on Artificial Neural Networks (ESANN), pp. 633–638 (2018)
Gallicchio, C., Micheli, A.: Echo state property of deep reservoir computing networks. Cogn. Comput. 9(3), 337–350 (2017). https://doi.org/10.1007/s12559-017-9461-9
Gallicchio, C., Micheli, A.: Why layering in RNN? A DeepESN survey. In: Proceedings of the 2018 International Joint Conference on Neural Networks (IJCNN), pp. 1–8. IEEE (2018)
Gallicchio, C., Micheli, A., Pedrelli, L.: Deep reservoir computing: a critical experimental analysis. Neurocomputing 268, 87–99 (2017). https://doi.org/10.1016/j.neucom.2016.12.089
Gallicchio, C., Micheli, A., Tiňo, P.: Randomized recurrent neural networks. In: 26th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2018), pp. 415–424. i6doc.com publication (2018)
Gallicchio, C., Micheli, A.: Deep echo state network (DeepESN): a brief survey. arXiv preprint arXiv:1712.04323 (2017)
Gallicchio, C., Micheli, A.: Deep reservoir neural networks for trees. Inf. Sci. 480, 174–193 (2019). https://doi.org/10.1016/j.ins.2018.12.052
Gallicchio, C., Micheli, A., Pedrelli, L.: Design of deep echo state networks. Neural Netw. 108, 33–47 (2018). https://doi.org/10.1016/j.neunet.2018.08.002
Gallicchio, C., Micheli, A., Silvestri, L.: Local lyapunov exponents of deep echo state networks. Neurocomputing 298, 34–45 (2018). https://doi.org/10.1016/j.neucom.2017.11.073
Grigoryeva, L., Ortega, J.P.: Echo state networks are universal. Neural Netw. 108, 495–508 (2018). https://doi.org/10.1016/j.neunet.2018.08.025
Hajnal, M.A., Lőrincz, A.: Critical echo state networks. In: Kollias, S.D., Stafylopatis, A., Duch, W., Oja, E. (eds.) ICANN 2006. LNCS, vol. 4131, pp. 658–667. Springer, Heidelberg (2006). https://doi.org/10.1007/11840817_69
Jaeger, H.: The “echo state” approach to analysing and training recurrent neural networks - with an erratum note. Technical report, GMD - German National Research Institute for Computer Science (2001)
Jaeger, H.: Short term memory in echo state networks. Technical report, German National Research Center for Information Technology (2001)
Jaeger, H., Haas, H.: Harnessing nonlinearity: predicting chaotic systems and saving energy in wireless communication. Science 304(5667), 78–80 (2004). https://doi.org/10.1126/science.1091277
Jaeger, H.: Discovering multiscale dynamical features with hierarchical echo state networks. Technical report, Jacobs University Bremen (2007)
Kawai, Y., Park, J., Asada, M.: A small-world topology enhances the echo state property and signal propagation in reservoir computing. Neural Netw. (2019). https://doi.org/10.1016/j.neunet.2019.01.002
Lukoševičius, M., Jaeger, H.: Reservoir computing approaches to recurrent neural network training. Comput. Sci. Rev. 3(3), 127–149 (2009). https://doi.org/10.1016/j.cosrev.2009.03.005
Mackey, M.C., Glass, L.: Oscillation and chaos in physiological control systems. Science 197(4300), 287–289 (1977). https://doi.org/10.1126/science.267326
Pascanu, R., Gulcehre, C., Cho, K., Bengio, Y.: How to construct deep recurrent neural networks. arXiv preprint arXiv:1312.6026v5 (2014)
Rodan, A., Tino, P.: Minimum complexity echo state network. IEEE Trans. Neural Networks 22(1), 131–144 (2011). https://doi.org/10.1109/TNN.2010.2089641
Rodan, A., Tiňo, P.: Simple deterministically constructed cycle reservoirs with regular jumps. Neural Comput. 24(7), 1822–1852 (2012). https://doi.org/10.1162/NECO_a_00297
Strauss, T., Wustlich, W., Labahn, R.: Design strategies for weight matrices of echo state networks. Neural Comput. 24(12), 3246–3276 (2012). https://doi.org/10.1162/NECO_a_00374
Triefenbach, F., Jalalvand, A., Schrauwen, B., Martens, J.P.: Phoneme recognition with large hierarchical reservoirs. In: Advances in Neural Information Processing Systems, pp. 2307–2315 (2010)
Verstraeten, D., Schrauwen, B., d’Haene, M., Stroobandt, D.: An experimental unification of reservoir computing methods. Neural Netw. 20(3), 391–403 (2007). https://doi.org/10.1016/j.neunet.2007.04.003
Weigend, A.S.: Time Series Prediction: Forecasting the Future and Understanding the Past. Routledge, Abingdon (2018)
White, O.L., Lee, D.D., Sompolinsky, H.: Short-term memory in orthogonal neural networks. Phys. Rev. Lett. 92(14), 148102 (2004). https://doi.org/10.1103/PhysRevLett.92.148102
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
A Selected Hyper-parameters
A Selected Hyper-parameters
Table 2 reports the DeepESN hyper-parameters selected by model selection for the experiments reported in Sect. 4. The reported values are the following: spectral radius \(\rho \), input scaling \(\omega _{in}\), inter-layer scaling \(\omega _{il}\), and number of layers L. We recall from Sect. 4.1 that the values of \(\rho \) and \(\omega _{il}\) are shared by all the layers. The selected hyper-parametrization for (shallow) ESN, are given in Table 3, where we report the chosen values of \(\rho \) and \(\omega _{in}\). We also recall from Sect. 4.1 that the total number of reservoir units is set to 500 for both DeepESN and ESN. While in the latter case all the 500 units form a single recurrent layer, in the former they are evenly distributed across the layers in the deep reservoir.
Interestingly, from Table 2 we can observe that constrained reservoir topologies in DeepESNs generally tend to show smaller values of the spectral radius and a deeper architecture than basic (i.e., sparse) reservoir settings. Comparing Tables 2 and 3 we also note that the values of spectral radius and input scaling selected for DeepESN and ESN correspond quite well in all the analyzed reservoir settings.
Rights and permissions
Copyright information
© 2019 Springer Nature Switzerland AG
About this paper
Cite this paper
Gallicchio, C., Micheli, A. (2019). Reservoir Topology in Deep Echo State Networks. In: Tetko, I., Kůrková, V., Karpov, P., Theis, F. (eds) Artificial Neural Networks and Machine Learning – ICANN 2019: Workshop and Special Sessions. ICANN 2019. Lecture Notes in Computer Science(), vol 11731. Springer, Cham. https://doi.org/10.1007/978-3-030-30493-5_6
Download citation
DOI: https://doi.org/10.1007/978-3-030-30493-5_6
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-30492-8
Online ISBN: 978-3-030-30493-5
eBook Packages: Computer ScienceComputer Science (R0)